NumPI

numpi@lists.dm.unipi.it

2 participants
265 discussions

Optimization & numerical analysis seminars: A. Frangioni
by Federico Poloni 19 Mar '15

19 Mar '15

In the next months we are holding a small cycle of joint seminars between the optimization and numerical analysis groups; their purpose is describing each other's research work, with an eye to possible new common areas of interest. Prof. Frangioni's seminar is the first one of the series. Time: Thursday 2015-Mar-19 (next week), h. 10:00 Place: Sala Seminari Ovest, Dipartimento di Informatica Speaker: Antonio Frangioni, Università di Pisa Title: *Iterative approaches for graph-structured linear systems arising in optimization* Abstract: When applying Interior Point (IP) algorithm to the solution of linear or convex separable optimization problems with strong graph structure, like the Min Cost Flow (MCF) problem, one has to repeatedly solve linear systems with strong graph structure. One of the two possible approach leads to systems with a M-matrix that is basically the weighted Laplacian of the underlying graph, where the weights (but not the graph) change at each iteration of the IP methods and typically become "very imbalanced" at the last ones. For this problem we have used two different techniques: Preconditioned Conjugate Gradient (PCG) methods, based on the idea of extracting a proper triangulated subgraph of the original graph which strictly contains a spanning tree. We have defined a new class of triangulated graphs, called Brother-Connected Trees (BCT), and discuss some fast heuristics for finding BCTs of "large" weight, starting from either the Kruskal algorithm or the Prim one. In addition we have studied multi-iterative techniques where the above PCG approach has been combined with specialized coarser-grid operators which preserve the graph structure of the projected matrix at the inner levels and smoothers, that have been completely characterized. Quite a lot of work still remains to do for understanding which is the best combination of preconditioning and coarsening at different stages of the overarching IP algorithm, implementing these techniques in an efficient solver, and extending them to "truly" interesting graph-structured problems like Multicommodity MCF ones. Everyone is welcome! -- --federico poloni Dipartimento di Informatica, Università di Pisa http://www.di.unipi.it/~fpoloni/ tel:+39-050-2213143

1 1

Seminario del Prof. J. Gondzio 20/3 a Firenze (RETTIFICA AULA)
by margherita porcelli 14 Mar '15

14 Mar '15

Carissimi, vi ricordo il Seminario del Prof. Jacek Gondzio (School of Mathematics, University of Edinburgh) di venerdì prossimo (20/3) a Firenze e, a rettifica di quanto comunicato in precedenza, preciso che il seminario si terrà presso il PLESSO DIDATTICO MORGAGNI (Aula 223). ********************************* Ven 20 Marzo, ore 14:00 Luogo: FIRENZE aula: Aula 223, PLESSO DIDATTICO MORGAGNI Viale Morgagni 40, Università di Firenze TITOLO: "Preconditioners for higher order methods in big data optimization" ABSTRACT: We address efficient preconditioning techniques for the second-order methods applied to solve various sparse approximation problems arising in big data optimization. The preconditioners cleverly exploit special features of such problems and cluster the spectrum of eigenvalues around one. The inexact Newton Conjugate Gradient method excels in these conditions. Numerical results of solving L1-regularization problems of unprecedented sizes will be presented. J. Gondzio joint work K. Fountoulakis ********************************* Cordiali saluti, Margherita Porcelli -- Margherita Porcelli Dipartimento di Matematica Università di Bologna Piazza di Porta S. Donato, 5 40127 - Bologna - Italy phone : +39-051-2094423 home page: http://web.math.unifi.it/users/porcelli ---------------------------------------------------------------- This message was sent using IMP, the Internet Messaging Program.

1 0

Seminari del Prof. Jacek Gondzio (University of Edinburgh)
by Margherita Porcelli 04 Mar '15

04 Mar '15

Segnalo 2 seminari che il Prof. Jacek Gondzio (School of Mathematics, University of Edinburgh) terrà nel corso della sua visita presso il Dipartimento di Ingegneria Industriale (UNIFI) e il Dipartimento di Matematica (UNIBO). ********************************* Ven 13 Marzo, ore 11:00 Luogo: BOLOGNA aula: Seminario I, Dipartimento di Matematica, Università di Bologna TITOLO: "Preconditioners for higher order methods in signal reconstruction" ABSTRACT: We address efficient preconditioning techniques for the inexact second-order methods applied to solve various sparse approximation problems arising in signal/image reconstruction. The preconditioners exploit two features of such problems: (i) sparsity of the solution, and (ii) near-orthogonality of the matrices involved. The latter originates from the restricted isometry properties frequently assumed in such applications. Spectral analysis of the preconditioners and their practical efficiency when solving linear systems in the Newton Conjugate Gradient method will be presented. If time permits then a few comments on the some other problems originating from the "Big Data" buzz will also be given. J. Gondzio joint work with I. Dassios, K. Fountoulakis and P. Zhlobich ********************************* Ven 20 Marzo, ore 14:00 Luogo: FIRENZE aula: Aula Tricerri, Dipartimento di Matematica e Informatica 'Ulisse Dini' Università di Firenze TITOLO: "Preconditioners for higher order methods in big data optimization" ABSTRACT: We address efficient preconditioning techniques for the second-order methods applied to solve various sparse approximation problems arising in big data optimization. The preconditioners cleverly exploit special features of such problems and cluster the spectrum of eigenvalues around one. The inexact Newton Conjugate Gradient method excels in these conditions. Numerical results of solving L1-regularization problems of unprecedented sizes will be presented. J. Gondzio joint work K. Fountoulakis ********************************* Cordiali saluti, Margherita Porcelli -- Margherita Porcelli Dipartimento di Matematica Università di Bologna Piazza di Porta S. Donato, 5 40127 - Bologna - Italy phone : +39-051-2094423 home page: http://web.math.unifi.it/users/porcelli

1 0

Reminder: Numerical Analysis Seminar, S. Simsek, 2014-09-23
by Federico Poloni 23 Sep '14

23 Sep '14

## Numerical Analysis Seminars, 2014-2015 Title: The matrix nearness problem for some special type of matrices associated with linear matrix equations Speaker: Sinem Simsek, Sakarya University and Kırklareli University, Turkey Date: Tuesday, 2014-Sep-23 Time: 14:00-15:00 Place: Sala Seminari, Dipartimento di Matematica Everyone is welcome! -- --federico

1 0

[Numpi.di] Numerical Analysis Seminar, S. Simsek, 2014-09-23
by Federico Poloni 18 Sep '14

18 Sep '14

1 0

Jump to page:

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

NumPI